Who is behind this

The Phone Path is written by an audiophile who spent years working through the hobby — from budget Chinese IEMs through flagship planar magnetics and multi-thousand-dollar DACs — and eventually found that a professional studio interface and a pair of Beyerdynamic headphones told him more about his music than anything he had owned before.

This site exists to share that perspective honestly, without affiliate commission pressure, manufacturer relationships, or the need to generate controversy for clicks. The calculator is free, the opinions are genuinely held, and the guide reflects both the documents behind the database and real listening experience.

How this was actually built, and by whom

I should be straight about the method, because the scale of it is not something one person does by hand in evenings. The database holds 280 sources and 578 headphones, and every figure in it is meant to be traceable to a document the manufacturer published. It is worth saying where that promise is not kept: on 23 headphone entries and 8 sources the only address is a shop, a distributor or a review site, because for those brands no maker document exists to find. Those pages say so in as many words, and the count is on the method page. Finding, reading and transcribing those documents — product pages, manual PDFs, spec sheets, the German edition of a data sheet when the English one drops the reference unit — was done with AI assistance, at a volume I could not have reached alone.

What is mine is the rules. What counts as a source. That a power figure with no stated load is not a specification. That an open-circuit voltage cannot become a power rating. That a sensitivity whose unit the maker never printed gets shown as printed and flagged rather than silently converted. That a product with no traceable figure comes out of the database instead of being filled in from a retailer's listing. Those decisions are the site, and they are mine.

That method is good at volume and consistency. What it is bad at is knowing when it is wrong. So the site is built to be checked rather than trusted: every figure carries a label saying whether it was published or worked out, every model page links the document it came from, and the whole dataset downloads as CSV and JSON so you can run your own arithmetic over it. That is not modesty — it is the only honest way to publish a number you did not measure yourself.

Mistakes found so far

Repeated passes over this material turned up real errors, and the worst of them were in my own reasoning rather than in the data:

• I claimed Audient had put a peak voltage into an RMS power formula. That cannot be shown: V_pk²/R and 2·V_rms²/R are identically the same number, so the arithmetic cannot separate a formula slip from both channels being summed. The page now states the factor of two and refuses to name a cause.
• I claimed two of Apogee's published figures could not both be true. They can — they are two different conditions printed on one line. Retracted.
• I claimed Beyerdynamic's single 96 dB across three impedances could not be a per-milliwatt figure. That is backwards: efficiency per milliwatt is roughly what stays constant between winding variants of one driver, so it is weak evidence for the per-milliwatt reading.
• The home page printed the Sennheiser HD 600 as needing 2.0 mW. The figure from this site's own data is 6.6 mW — the 2.0 came from reading Sennheiser's per-volt sensitivity as if it were per-milliwatt, which is exactly the error this site exists to explain.
• The HiFiMAN Sundara was listed at 37Ω and 94 dB. Those are the figures HiFiMAN printed for the original 2018 model; its current product page and its February 2025 owner's guide both say 32Ω and 92 dB. The database now holds what the maker publishes today, and the model page says which revision that is. Neither revision states a reference unit.
• A reader pointed out that Sennheiser's 97 dB (1 V) for the HD 600 sits six decibels below the 103 dB (1 V) Sennheiser prints for the HD 650 and HD 6XX, which are the same impedance and the same driver family, and that independent measurement puts the two together. He is almost certainly right. The published figure still stands on the page, because that is what Sennheiser prints — but the page now says the maker's own line-up contradicts it, and links the evidence. The same treatment went to the DT 770, where one 96 dB covers the 32, 80 and 250Ω versions.
• The JDS Labs Element III MK2 was shown at 1.3 W into 32Ω when it can deliver 2.3 W. JDS publishes both, as Standard and Boosted modes of the same output, and the build took the lower of the two — safe by default, wrong here. It understated the amplifier by 2.5 dB on every page it appeared on.
• Two product descriptions inherited from the first version of this database contradicted the research sitting next to them: the Element III MK2 was called “fully balanced” when it has no balanced output at all, and the Topping A30 Pro was said to publish balanced figures when Topping's own comparison table classifies that amplifier as single-ended and leaves its balanced power blank. The numbers had been traced; the sentences had not.
• Thirty-seven figures were counted as coming from a manufacturer on the brand pages while the product pages said, correctly, that they were unverified. Two pages of this site were saying different things about the same number.
• A cell labelled “published at this exact load” showed 7.4 mW where Audient had printed 7.35. The route from milliwatts to volts and back, plus rounding, had quietly edited a figure the page promised was untouched.
• The lists that check named claims against the database were checking themselves in one direction only. A number could be changed in the prose and the check still passed, because it compared the list with the data rather than the page with the data. It now requires the figure to appear on the page as well.
• Twenty-six headphone entries displayed a sensitivity that contradicted the provenance note printed on the same page. All of them now agree with the document.
• The output-impedance page listed reference figures under the words “all from the makers' own published figures”, and two of them were not: the Apple USB-C dongle, for which Apple publishes no electrical specification at all, and the Hidizs S9 Pro, whose figure this site's own model page marks as unverified. The list is now built from the database and refuses any entry that does not pass the published-by-the-maker test, so the sentence cannot come apart from the list again.
• The converter page named Focal as a maker that quotes sensitivity per volt. Focal quotes per milliwatt, on fourteen of the fifteen models here. The sentence had been written from memory and never checked against the database sitting underneath it. Both lists on that page are now counted from the data, over figures that come from the makers' own documents.
• The download page stated the size of the CSV as the length of the text it had just built, not the size of the file it had just written. The BOM and the line endings add over five hundred bytes, so the figure was a kilobyte light. Measured from the file now.
• The line Google shows under the spec-sheet audit said “173 models state sensitivity per milliwatt, 82 per volt, and 137 print a number you cannot use”. The page itself said 254, 139 and 174. The description had been typed once and never touched again while the database grew, and nothing checked it, because every check here read the visible text and no check read the metadata. It is computed from the database now, and the checker reads descriptions too.
• Sixteen brand pages overstated how much of their data came from the maker. The share counted a figure as the maker's whenever the maker's unit was printed beside it — which is a different question from who printed it. BLON and JadeAudio therefore read 100% while not one of their figures came from either company, and a reader clicking through to the model page found it saying exactly that. A record standing only on a shop page now counts as not traceable, which puts BLON, JadeAudio, 7Hz and Thieaudio at 0% and Simgot at 29%. The share is also recomputed from scratch in the checker now, so a change to the rule in one place cannot slip past it.
• The Simgot EA500 page said “Simgot publishes this as 123 dB/Vrms”. Simgot does not. Its own product page prints “123dB” with a frequency and no reference unit at all; the “/Vrms” is on the shop pages, and this site had taken it from one of them. The entry now shows the figure as Simgot printed it, which makes the per-milliwatt reading implausible — and the page says so, with the per-volt reading of 105.0 dB/mW beside it. Two of its siblings, the EW200 and EM6L, did check out on Simgot's own pages and have moved off the shop listings.
• Eight sources were in the same position, and only one of them was flagged: the figures came from Audiophonics, Apos, a dealer, a review site or a manual aggregator, with no address on the maker's own site. Those eight pages now say so too. It is the same test on both halves of the database, run from one list of domains.
• Thirty-one headphone entries had no manufacturer document behind them at all. Their only source was a shop — Linsoul, HiFiGo, Concept Kart, a German distributor, in one case a review blog — while the page above the link still read “Simgot publishes this as…”. Twenty-two of them carried a warning; the other nine did not, because the warning was set by hand and hand missed some. It is now worked out from the source address itself, every one of the thirty-one says which shop the figure came from, and a source domain that is on neither the maker list nor the shop list stops the build instead of slipping through as a manufacturer.
• The retraction above did not reach the pages it was about. The site kept telling readers of the three DT 770 pages that a single 96 dB “cannot be the per-milliwatt sensitivity of all three” — the exact sentence retracted here, and the opposite of what the audit page and the impedance-versions page say. Those three boxes now say what the shared figure really shows: it rules out the per-volt reading and leaves the per-milliwatt one standing, without Beyerdynamic ever having said which it is. A retraction that only lives on the about page is not a retraction.
• On twenty-nine sources this site said an output impedance was “not published” while the maker's page did show an ohm figure — for the line output, or as the range of headphones to plug in. The database was right to leave the field empty and the pages were right in substance, but a reader holding the spec sheet had no way to tell that from a mistake. Each of those product pages now quotes the figure and says what it belongs to.

If you find another, please tell me. A correction is worth more to me than a compliment, and the manufacturer's document always beats my spreadsheet.

Contact: temirowski@gmail.com